Find ID numbers, phone numbers, card numbers or names in your sources and remove them before an agent uses them. The AI suggests, you approve.
A clinic that uploads old patient emails, or a lawyer who uploads signed contracts, may not want ID numbers, card numbers or private phone numbers inside a notebook that an agent answers from. Redaction finds them and replaces them in the source text. It works in your notebooks, on one source, on several at once, or on every new upload.
1. Click the shield on the source: Open the notebook. On the source's row click the shield button Redact sensitive data. A window called Redact Sensitive Data opens.
2. Choose how to search: Click AI: recommend what to filter to let the AI suggest, or describe what to remove, for example "ID numbers and phone numbers", and click Find by instructions.
3. Check the findings: Each finding shows its category, a confidence level, the text and what it will be replaced with. Untick what should stay. Some findings say Needs manual review when the exact text was not found; check those yourself.
4. Choose what happens to the original: Under What to do with the original? choose Delete permanently (default) or Keep original for undo. Then click Apply Redaction. Deleting asks you to confirm with Yes, purge & apply.
Delete permanently cannot be undone: With the default choice the original text is deleted for good. Choose Keep original for undo if you may need it back: the original is kept so the redaction can be reverted later, and after a scan of several sources Revert (restore original) brings it back.
In the notebook's Chat tab, select two or more ready sources in the source list. The button Scan & Redact Selected appears with the number you picked. Click it, then Start scan. Each source shows Queued, Scanning, Done, Error or Skipped; when the scan ends, review the findings per source and apply.
1. Open the policy: At the top of the open notebook click Redact. The window is the auto-redaction policy for this notebook.
2. Turn it on: Tick Auto-redact every newly uploaded source. Under Mode choose AI recommend or By instructions and write the instructions. Set Minimum confidence (0.75 by default): findings below it are left alone. Choose whether to keep the original, then Save.
You can do the same from an AI tool connected to your account: ask it to scan a notebook for sensitive data, show you the findings and apply what you approve. See Connect AI tools.
Q: Does redaction change the file I uploaded?
A: It changes the source text the notebook keeps and answers from. Download or keep your own copy of the original file if you need it.
Q: Will an agent still see the removed text?
A: No. After redaction the notebook answers from the redacted text. With Keep original, the original is kept only so you can revert.
Q: The AI missed something.
A: Run it again with instructions that name what to find, for example "the patient names in the greeting line", and lower the minimum confidence for automatic redaction.
Q: Does it cost credits?
A: Finding sensitive data is AI work on your account, like other AI work.
Related: Notebooks, folders and uploads, Ask the Brain, What the Business Brain is, Agent permissions: what each agent may do